A visual work created by an artificial intelligence (AI) acting without human supervision cannot be protected by copyright in the United States, confirmed a US federal judge, Friday August 18. The latter was deciding on the validity of a previous verdict, issued in early 2022 by another American body, the United States Copyright Office.
This institution, based in Washington, had refused to register in its directory of protected works an image generated by Dabus, an AI designed by Stephen Thaler. This scientist presents the image, titled A Recent Entrance to Paradise, as a standalone creation of Dabus: no human instructions would have weighed on the final result. A radical approach, which could have come from a conceptual artist, but which germinated in the mind of this researcher and founder of Imagination Engines, a company whose stated scientific objective is to manufacture conscious and sensitive AI. An AI should also be recognized “as an author”, argued Stephen Thaler’s lawyer to the court.
“Without tutelage of a human hand”
In its judgment, the Federal Court recalls that American law has never “protected works generated by new technologies operating without the supervision of a human hand”, and that, according to well-established case law, it is human creativity that “founds the possibility of receiving a copyright”. According to her, the purpose of copyright law is to “encourage humans to engage” in creation.
However, this judgment does not condemn all visual works generated by AI. When it is used as a tool for assisting creativity, the person who has used it may, under certain conditions, manage to obtain copyright, if we rely on a notice from the Office of copyright notice published in March 2023. The office then argued that works where the making of an AI was selected and arranged “creatively enough for the work produced to constitute an original work of authorship” could be covered. , case by case. This opinion is certainly not binding, and it comes quite early in the legal debate that is beginning, but it seems consistent with case law whose foundations were laid in the 19th century following the emergence of the photography.
Neither this opinion nor the judgment of the federal court in Columbia answers another question: can AIs be trained on the basis of millions of visual and textual creations drawn from the Internet? Another thorny debate that will probably take a few years to be finally settled by the courts. Unless the political powers decide to set new rules of the game.